Introducing Nova - A Deterministic C++ Logging Library With Domain-Based Routing

Repository: https://github.com/kmac-13/nova/
Benchmarks: https://github.com/kmac-13/nova/blob/main/docs/BENCHMARKS.md

I am pleased to announce the initial release of Nova - a modern C++ logging library focused on deterministic behavior, compile-time configurability, and flexible domain-based routing for systems ranging from hosted platforms down to bare-metal and safety-critical environments.

Why Another Logging Library?

There are already several quality C++ logging libraries available. However, most logging libraries organize routing and filtering around severity levels and rely on global logger configuration or runtime string-based logging categories. Engineers are often forced to encode subsystem behavior into a limited set of severity levels while also considering which thresholds will be enabled in production. This also leads to situations where enabling debug logging for one subsystem effectively requires enabling debug logging across unrelated areas of the application.

Nova instead treats logging domains as compile-time types, allowing logging configuration and routing to directly reflect application structure rather than forcing subsystems into global severity categories.

Domains can represent subsystems, modules, interfaces, classes, libraries, or any other domain-specific concept, and each domain can be independently enabled, disabled, or routed without reliance on shared global configuration. Because domains are independent types rather than shared string identifiers, libraries can define their own logging domains without interfering with application or third-party logging configuration.

• type-defined logging domains - log against subsystems, modules, classes, libraries, or any other concept
• compile-time elimination of disabled domains - language-guaranteed in C++17, optimiser-dependent in C++11/14
• compile-time routing - avoids reliance on global logger registries or shared runtime configuration

Additional goals of the library include:

• simple pipeline and extensible API
• deterministic behavior - no heap, no exceptions, no RTTI
• support for platforms ranging from hosted systems down to bare-metal
• fast enough for demanding real-time and multithreaded workloads

Nova also includes Flare, an async-signal-safe crash and forensic logging component that writes structured diagnostic records directly to disk from signal handlers - without heap allocation, locks, or non-signal-safe C++ runtime features.

Example

#include <nova.h>

// define a domain (can be any type)
struct MotionPlanner {};  

// configure the domain with a name (MOTION), enabled state (true), and clock type (steadyNanosecs)
NOVA_LOGGER_TRAITS( MotionPlanner, MOTION, true, kmac::nova::TimestampHelper::steadyNanosecs );

int main()
{
    // configure motion planner sink as mpSink
    ...

    // bind the mpSink to the MotionPlanner logging domain
    kmac::nova::ScopedConfigurator config;
    config.bind< MotionPlanner >( &mpSink );

    // log
    NOVA_LOG( MotionPlanner ) << "Planning trajectory...";
}

Here we can see that the MotionPlanner domain is defined, the traits for the domain are configured, a target sink is bound to the domain, and logging is performed. In this example, the domain is a simple, empty struct, but a domain can be any type, including interface, abstract, or concrete classes. A domain can even be a specific class, and logging can be limited to the scope of that class.

Using types as logging domains enables compile-time routing, strong subsystem separation, and per-domain configuration and enablement. Disabled domains can be eliminated entirely by the compiler, and type names prevent the silent runtime failures that string-based routing can introduce.

Additionally, per-domain control means that enabling verbose logging for one subsystem has no effect on any other - there is no shared severity threshold to raise or lower across the entire application just to see detailed output from a single area.

Performance

Nova has been benchmarked against several popular C++ logging libraries, including Quill and spdlog, using:

• multiple thread counts
• fixed queue sizes
• sustained throughput scenarios
• guaranteed-delivery latency tests
• both real file sinks and counting sinks to isolate I/O overhead

The benchmarks intentionally normalize queue sizing and backend threading models to avoid structurally advantaging any particular library configuration.

Results varied by workload, but several patterns consistently emerged:

• Nova performed particularly well in guaranteed-delivery scenarios and bounded-memory configurations
• Nova's synchronized sink significantly outperformed comparable synchronous configurations from other libraries
• Nova's async backend (using a no-I/O counting sink) demonstrated competitive throughput across thread counts while maintaining zero per-record heap allocation throughout
• Quill achieved extremely high front-end enqueue throughput, but with substantial drop rates under sustained overload conditions
• spdlog async showed performance degradation under heavy multi-threaded workloads

While Nova does not always achieve the highest theoretical front-end enqueue rate, it performed extremely competitively across a broad range of realistic workloads, especially where deterministic behavior and bounded memory usage are important.

Full benchmark methodology and raw benchmark data are available in the repository.

Design Philosophy

Current Status & Feedback

The initial release is now available at the repository linked above. I am still actively working on improving the library with additional features such as:

• additional sinks/backends
• packaging and integration polish
• binary logging

I would appreciate feedback on any aspects of Nova (e.g. integration experience, cross-platform/compiler issues, feature requests). If you try Nova in a project, I’d love to hear how it performs and where it can improve.

Thanks for reading.